Few Ideas for Rightholders to Share the Benefits of Generative Artiticial Intelligence in the Field of Intellectual Creation
From the Three-step-test to the the Domaine Public Payant
Few Ideas for Rightholders to Share the Benefits of Generative Artiticial Intelligence in the Field of Intellectual Creation[1]
Se vogliamo che tutto rimanga come è, bisogna che tutto cambi[2]
From contributive AI to substitutive AI? Artificial intelligence, a tentacular phenomenon with blurred contours, is the focus of much attention, especially now that the public has discovered the potential of so-called « generative »[3] artificial intelligence, which paves the way for students to stop writing their own papers, and for novelists, musicians, choreographers and painters to delegate to machines the task of creating « works » that they have generated with the help of « prompts ». For the time being, creators are the first ones to use these tools to help them in the process of creating or executing their works, this being the fruit of several millennia of man-machine collaboration (cavemen already used tools to sculpt). But these prompts could in turn be generated by the machine, without the user having to formulate a question, a wish, a pitch, a synopsis, whatever you want to call it, so that the figure of the author might tend to disappear, or at least diminish substantially. There are offers on the market for the sale of pre-written prompts, as well as new solutions that allow, for example, the generation of an image from music, and the invention of all kinds of instructions to feed the machine.
Already, certain press articles are automatically produced from news, reducing the workload of journalists. Translators, dubbing actors and even scriptwriters clearly see the threat of the great replacement by new tools looming over them, and are seeking various ways to delay the outcome. The industry is following suit. A video-game studio has just created a virtual band featuring characters from a game whose success will no doubt be assured. A major record label is signing virtual artists in a bid to make substantial savings. Even if this is only producing “can-music”, it deprives those who, as human beings, had been so far responsible for this kind of creation. « Now and Then », the Beatles’ last track, produced with the help of voice-cleaning tools on John Lennon’s voice, opportunely resurrected, is already a hit[4]. Maginot lines won’t hold for long, and it’s a safe bet that the self-limitation of the use of generative machines on ethical grounds won’t overcome the unbridled competition between operators.
In the short term, it’s fair to ask whether it will be worth being a human author, when the machine, thanks in particular to the abundance of data used to train its models, is likely to outshine her creative capacity, at lower production costs[5]. The transition from argentic to digital photography already illustrated this Schumpeterian phenomenon in the 2000s, and many creative professions have disappeared or become scarce due to technological « progress ». And yet, because of their rapid evolution, the productions generated by machines using these models are likely to siphon off market share from human creation.
Sharing the value. This is not the place to discuss the extent to which these assertions are fantasy or reality. If my conviction is that art is not simply a mill of the world’s memory; that the mechanical fertilization of works based on individual desire in no way guarantees their social interest and that, consequently, there will always remain space for human creation, even in the presence of new objects generated by or with AI, the movement is underway and will no doubt leave aside a certain number of tasks usually carried out by natural authors. The developers of IT solutions, who are already dominant economic players, will reap the benefits of this tectonic shift, once paid-for remuneration services have been set up, as it is currently the case.
Legislative wavering. Lawmakers, for their part, are caught in a vice: limiting the potential of AI and risking the hindering of innovation or authorizing it and speeding the downfall of existing players. France’s timid stance on AI regulation bears witness to this stalemate. The issue of value transfer between creators on the one hand and the content distribution industry on the other, which was at the heart of discussions when the 2019 directive on copyright in the digital single market was adopted, was only partially anticipated. By focusing on Article 17 on the use of works by platforms, rightsholders underestimated the threat posed by the exception allowing protected content to be “harvested” by text and data mining processes. Thanks to the lobbying of the IT industry, which wanted to have a free hand in the development of AI, this exception to the exclusive right was finally granted without monetary compensation, subject only to an « opt out » procedure obliging rights holders to use machine-readable tools to signal to AI players their wish to prevent them from carrying out this search. Upstream, the absence of monetary compensation for the use of data, notably consisting of protected works or other objects, deprives traditional players in the value chain of remuneration, while downstream, they suffer from the risk of their creations being crowded out by these attractive products.
Seeking for valuable legal and economic solutions for outputs. On the downstream side, proposals are flourishing for either the recognition of a copyright[6] or neighbouring right for the benefit of the machine designer[7] or user, or the introduction of a tax or other levy system on the use of machine-generated works. Some countries, such as the United Kingdom, propose a hybrid system whereby machine-generated productions are granted a reduced form of protection. Models exist within intellectual property law to counter a market failure, i.e. a factual impossibility of controlling usage and reaping the benefits; it is possible, for example, to introduce a remunerated exception. The use of content generated by machines protected by intellectual property rights could trigger such a mechanism. The model of fair compensation for private copying offers advantages in this respect, not least the collectivization of the mechanism, which means that it can be borne by almost everyone. However, the theoretical question of the basis for compensation is scarcely resolved, if we tend to consider – as is the option adopted by the present article – that content generated a priori by machines should not be protected by copyright, since a human has not participated substantially in the composition of the work. If copyright is denied to such creations, it is hard to see why a royalty should be levied on their use, even if it competes with human works, based on an exception… to copyright.
Would the regime inspired by orphan works be more appropriate? When, after diligent, proven and serious research, no rights holder can be found (article L. 135-3 of the French Intellectual Property Code), because their trace has been lost, the law provides for the possibility of exploiting the work within the framework of a remuneration system administered by a collective management organization and designed to prevent free use from competing with protected works. This interesting economic logic tends to protect the exploitation of protected works from unfair competition from works that would be accessible at a lower price. But, apart from the fact that the mechanism has not been as successful as expected, the problem of the legal basis for extending such a system of compulsory collective management arises when it has been designed for creations that are in principle protectable by copyright.
Perspectives. It’s a bleak picture, and many voices are being raised today to denounce this unbalanced situation and call for the system to include, at the very least, a chance of remuneration for « human » authors. The aim of this contribution is to consider the possibility of such « compensatory » mechanisms for mining practices, on the one hand, and for the exploitation of outputs generated by AI, on the other. As we shall see, however, the upstream and downstream sides necessarily converge in the search for balanced solutions. From a deliberately utilitarian perspective, the first stage is to consider the compatibility of a free TDM exception with the three-step test, when this mining ultimately serves to produce quasi-« works » by a generative AI (I.). The second stage is to assess the possibility of setting up a domaine public payant for to the outputs produced by a generative AI system (II.).
- The Inconsistency of the TDM Exception with the Three-Step test for Outputs Equipollent to Works
One of the first possible compensatory leverage would be to rethink the text and data mining exception, which currently may be used without any monetary compensation for rightholers, and to add one. To this end, arguments are being developed to contest the compliance of the proposed text and data mining exception provided for by Article 4 of the CSDM Directive, with the three-step test. However, it seems that arguments relating to the exception’s incompatibility with normal exploitation, or unjustified prejudice miss their target in part (1.1.) if they are not first contextualized in the light of the test’s first condition, namely the requirement of a special case which, for the generation of information equipollent to works by generative AIs, is lacking (1.2.).
- The non-relevant critique of the TDM exception as regards the last two conditions of the three-step test
- Opt out and normal exploitation
Opt out on the input. A first defense against the free use of content mined by AI systems is to activate the opt-out possibility provided for in the TDM exception, i.e. to express a decision of refusal of the reproduction/scrapping of the content and thus make protected works unavailable to data mining. This is what the SACEM[8] and other CMOs has just done, to signify that access to its repertoire will not be free but will be subject to negotiation. Opting out would therefore be the preliminary step to the return of exclusive rights and the possibility to benefit from associated sources of remuneration. The solution, attractive on the surface, nonetheless raises a number of issues, particularly as regards its generalization and implementation methods (what procedures? what tools?), but above all opens up the possibility that training will ultimately be carried out on datasets that do not contain the said works, or from countries that do not recognize protection, thus depriving AI of French or European repertoires, or depriving French AI players of development opportunities, whereas the United States, Japan or England have not envisaged such limits on TDM. Thus, from a purely economic point of view, the exercise of the opt-out is far from guaranteeing any chances of gains for rightholders, even if the prospect of monetary negotiation is of course favored by the latter.
Non-compliance with the three-step test of the TDM exception for generative AIs. Normal exploitation and opt-out. A second argument challenges the conformity of the TDM exception used for generative AI with the three-step test. Need we remind you that both international (Berne Convention, TRIPS) and European (Directive 2001/29) legislators require that the introduction of exceptions and their exercise shall comply with the three-step test? To be admissible, the exception must constitute a special case, it must not conflict with a normal exploitation of the protected object, and it must not unreasonably prejudice the legitimate interests of the rightholders. It is possible to argue that article 4 of CSDM, which introduces an exception for text and data mining, does not meet these three conditions.
However, the infringement of normal exploitation is not the easiest to demonstrate. Indeed, some authorized authors, such as Martin Senftelben[9] argue that, precisely, the ability to opt out, by putting owners in a position to reserve the market for themselves through the opposition mechanism, makes it possible to prevent such an infringement from occurring. Others argue that this processing works as data does not correspond to an exploitation of the protected object that « normally » enters the markets invested by the owners. However, this interpretation is questionable, insofar as it assumes that rightholders cannot enter new markets of this type, which nothing prohibits. It is also contradicted by the ability to opt out, that precisely includes data mining within the scope of the monopoly, as soon as the holder expresses his wish to be on this market.
- Is the prejudice unreasonable?
Absence of compensation and non-compliance of the TDM exception with the three-step test. Another way to contest the conformity of the exception with the test is the possible existence of an unreasonable prejudice caused by the TDM to rightholders, that the absence of economic compensation would demonstrate. It would amount to consider that for TDM, which is the pre-requisite processing to generate works-like outputs, the absence of remuneration would deprive unduly the right holders from a compensation for the harm they suffer because of the competition of these ersatz on the output market. The idea is an interesting one but may raise two types of objections: on the one hand, the compensation would be linked to the harm occurring on the downstream market and not on the “mining” market ; on the other hand, it is not completely in line with the recital 17 of the CSDM Directive according to which “in view of the nature and scope of the exception, which is limited to entities carrying out scientific research, any potential harm created to rightholders through this exception would be minimal. Member States should, therefore, not provide for compensation for rightholders as regards uses under the text and data mining exceptions introduced by this Directive”. The second sentence seems to exclude such compensation for both TDM exceptions within the Directive, namely TDM for purposes of scientific research (art. 3) and the exception or limitation for text and data mining (art. 4). The second objection may be overcome though, because the assumption that the harm created to the rightholders is minimal that justifies the absence of compensation only refers to the entities carrying out scientific research, namely the hypothesis of article 3. It is possible, consequently, through an a contrario interpretation to defend the idea that when the purpose is not scientific research and that the entities benefiting from the exception are not the organizations that carry out such research, the prejudice would not be minimal and, as the result, that a compensation would be justified. Furthermore, a monetary compensation was envisaged in former versions of the Directive before its adoption for the second TDM exception. Member States could, in this perspective, create a compensation system for the exception when the prejudice is not minimal.
Risks of this interpretation. This way of thinking is attractive but seems nevertheless unsatisfactory. First, it is always dangerous to rely on a contrario interpretation method, which proves to be sometimes unreliable. Indeed, it cannot be automatically deduced from this logic that when the TDM exception does not apply to research establishments, then it must give rise to compensation because the prejudice is not minimal. It would be fairer to open the way to a probability by formulating the hypothesis that, outside the scope of the exception of TDM for research purposes, the prejudice might not be minimal, which means that it would be possible to distinguish, within article 4, situations in which the prejudice would be potentially substantial from those in which it would be minimal. Not all mining operations lead to the generation of objects that are likely to cause unjustified prejudice to rightholders[10]. If, for example, the product of the TDM consists in tracing Proustian onomastics as part of an analysis of La Recherche du temps perdu by unconditional fans of the great Marcel, it is hard to see how such processing of a work (which is otherwise in the public domain) could justify compensation for right holders. The result of the search – in this case, the appearance of constants in the texts – is the very essence of TDM, and does not prejudice the rightholders, since it is akin to an intelligent analysis, not an exploitation of the work as such.
Furthermore, the compensatory mechanism can only be global, and the outputs may or may not be protected by intellectual property rights. Processing mechanisms mine colossal amounts of diverse information, and it seems very difficult to distinguish what falls within the scope of literary and artistic property protection among them. Calls for dataset transparency could, if they are heeded, help to identify the marginal share of protected content in the corpus searched, but it is likely that this share will then appear relatively minimal, and that the damage will consequently also be minimal[11][.Thus, neither certain forms of results, nor the share attributable to the owners of protected objects in them, ensure the existence of sufficient harm. While the possibility of non-conformity with the absence of unjustified prejudice cannot be ruled out, it must be contextualized and lead to a new approach, taking into consideration only the specific case of generative AI.
- The lack of conformity with the special case first condition of the test for AI-generated ouputs that are equipollent to works
Non-conformity with the three-step test of the TDM exception for generative AI. In my view, the most effective way to challenge the conformity of the three-step test for the text and data mining exception is to argue that it does not correspond to a special case when the purpose of the processing is to produce a piece of information which is similar to a work. Unlike the exception in article 3, which associates the mining with a research purpose – the mining is then the means – the exception in article 4 envisages the mining as being the purpose. As its name suggests, Article 4 introduces an exception « for » text and data mining. The European legislator has thus enshrined the possibility of reproducing or extracting, without prior authorization from rightholders, works and other protected objects that are lawfully accessible for the purposes of text and data mining. It is “mining for the sake of it” since mining is both the means and the end. This circular logic poses a problem of delimitation of the exception and, consequently, of its conformity with the first condition of the test, the existence of a special case.
What, in principle, is the scope of the exception? As mentioned above, according to Article 2(2) of the Directive, data mining is authorized “to make possible any automated analytical technique aimed at analysing text and data in digital form in order to generate information which includes but is not limited to patterns, trends and correlation« . However, the absence of a definition of the notion of information opens the door to all kinds of results, including the possibility of ultimately extracting outputs that are similar to « works ». However, to be useful, one need to distinguish between « information » that resembles a work or other protected object in terms of its characteristics or functionalities, and « information » that does not (raw information, mathematical formula, etc.). « Work[12] » means here creations that fall within the scope of intellectual property, but which may not be protected if they fail to meet the other conditions that trigger the application of exclusive rights.
The three-step test on the market related to the outputs that are substitutable to “works”. In this sub-hypothesis of the exception, the mining could consist of analyzing protected works aimed at generating work-like information. Following this reasoning you may come to the absurd conclusion that an automated analytical technique aimed at analysing a work can potentially generate information that is nothing other than the perfect reconstitution of the work itself. In this case, the analysis process would consist of generating an exact copy of the model, through an innovative technical means. In this extreme hypothesis, mining would appear as just another reproduction process, with no external purpose to justify it. It therefore seems possible to state that a process of mining works with the aim of generating work-like outputs would not meet the condition of a special case, since it would pave the way to any use of the work produced through this automated technique.
This proposition may raise two ranges of critiques. The first one is that the outputs generated by AI are not works. The second is that the result of mining itself is never to generate information that are similar to a work, since one or more intermediate algorithmic operations are necessary to achieve such a result. Regarding the first objection, it is possible to consider the relationship between the work and the information or data as bijective. If the work is treated as information insofar as it is the object of the mining, we could also consider that the piece of information produced by the mining could also be akin to a work, admittedly an unprotected « work », but a piece of information that takes the form of a work. As for the second, it can be dismissed on the basis of an overall functional vision of the treatment process.
Outputs/works or work-like outputs. Assimilating certain outputs to works is a theoretical step which, on reflection, is not so complicated to take. It involves drawing on the distinction, often overlooked in case law, between the notion of a work as a category potentially eligible for protection on the basis of its objective characteristics, and that of a protectable work which, by virtue of other criteria, meets the conditions for triggering legal status. The European Court of Justice has drawn a distinction between two criteria: intellectual creation and the need for sufficient precision in the expression of a form[13]. Thus, a work may fail to obtain protection because the judges consider that, despite belonging to the domain of « literary and artistic » creation, it does not present the other characteristics required to be protected: absence of creation by a natural person, absence of originality, etc. There are works whose legal status does not depend on the fact that they have been created by a natural person. Some works may not be protected because of this absence, but they do not lose their qualification as works for all that – they may not be works of the mind/spirit. In other words, an output generated by an AI could never acquire protection as regards copyright threshold, in the absence of decisive human intervention, without necessarily rejecting the assimilation of such productions to a « work » as regards its external features.
The hypothesis of an output generated by mining that would be equipollent to a work. Incidentally, it may not be necessary to use this qualification since what matters is the existence of a substitutive effect on the market. Even if it is not a « work », the output generated by the AI has the characteristics of a work in terms of form or potential uses and is therefore likely to replace the work with an ersatz, a substitute, as it may be called, which from an economic point of view, will be on the market of exploitation of the genuine human works. The problem posed by generative AI in the field of copyright does not primarily concern outputs that do not resemble works, but rather those that do. The application of the three-step test should therefore be restricted to this last hypothesis. Owners’ claims cannot go so far as to impose a prior authorization mechanism on any type of mining operation, whatever the result of the analysis should be, on the sole grounds that an act of reproduction occurs at some point in the processing chain[14] but they could be entitled to do so if the intended result of the service relying on the mining is it to produce work-like ouputs (such as Midjourney, Stablevision…).
Former rejection the monopoly on the “documentary processing” of works. Moreover, French case law has already refused to include within the scope of the reproduction right certain purely technical processing operations that have no impact on the market of the exploitation of works. This was already ruled a long time ago, in the famous Microfor rulings[15]. In the first case, the Cour de cassation ruled that article 40 of the law of March 11, 1957 (now article L. 122-4 of the French Intellectual Property Code) « does not apply to the publication, by any means whatsoever, of an index of works enabling them to be identified by keywords« ; nor did it apply to « a purely descriptive analysis carried out for documentary purposes, exclusive of a substantial statement of the content of the work, and not enabling the reader to dispense with recourse to the work itself« . The 1987 decision of the Assemblée plénière, for its part, stated in an opinion that « if the title of a newspaper or of one of its articles is protected in the same way as the work itself, the publication for documentary purposes, by whatever means, of an index containing the mention of these titles with a view to identifying the works listed does not infringe the author’s exclusive right of exploitation« . The Cour de cassation has thus solemnly established that indexing practices do not fall within the scope of the exclusive right, even if the addition by the Assemblée plénière of the condition of « publishing for documentary purposes » opens the possibility of nuancing the solution. It therefore seems difficult to argue that mining that actually led to the production of simple analysis information should systematically fall within the scope of the exclusive right.
“Competitive” test. What emerges from this case law is the existence of an underlying competitive test between the result of the analysis and the market for the work under analysis. For the judges, when the product of the analysis replaces consultation of the original work, by dispensing the reader from having to go there and by reproducing a « substantial » account of it, it falls within the scope of prior authorization. On the other hand, if it does not fulfil this substitutive role, it does not need to be authorized by the owner of the work being analyzed. It was on the basis of this reasoning that the Paris Court of Appeal rejected the application of copyright to image engines, since the result of the processing was a means of searching for an image[16] and the reproduction of images in thumbnail form did not exceed what was necessary for this function of the tool. It is true that, since then, Directive 2001/29 has given the reproduction right a very broad definition covering all forms of reproduction, albeit the compulsory exception for transient copying was precisely intended to limit the scope thereof for certain automated processes. It is worth noting that this exception, which may justify certain mining practices, establishes a condition of absence independent economic significance, which precisely allows a copy made in the course of a remunerated service, which places itself on a different market from that of the exploitation of the work itself[17].
Overall economic approach. As far as the technical objection is concerned, it is certainly admissible insofar as the mining processes carried out by an AI system, even a so-called generative one, presuppose a succession of actions and actors that do not fit in well with a global approach. There is a long way to go from the mining to the generation of content by AI, and several intermediate actions are needed to complete the operation, so that this preparatory mining would only be one stage in the final process of « generating » the output, from which it would have to be distinguished. It would seem, however, that this argument can be overcome through an « reversed » approach to the operation. In fact, the mining exception should only be considered from the point of view of copyright if the operator generating work-like content has been able to do so thanks to training data containing works or other protected objects. The economic operation and contractual engineering enable operations to be chained together and traced from one operation to the next, to ensure that a sufficient link exists between the upstream market (the mining) and the downstream market (the work-like output).
Distinction of the exception according to the destination of the mining. The final destination of the processing may differentiate two categories of regimes: TDM exception remains free when the AI tools’ analysis destination is to provide information that are not substitutable to works or other protected subject matters. If, on the other hand, the analysis is intended to generate work-like outputs, even if this generation presupposes subsequent operations, it does not correspond to a special case, as it fails to determine a precise framework of intervention, different from the exploitation of the work itself or its derived forms. If the condition of a special case is not met, the exception cannot be implemented, and we return to the exclusive right.
At the end of this demonstration, it is therefore possible to put forward the following hypothesis: the current mining exception cannot pass the three-step test in cases where the analysis carried out leads to the generation of work-like outputs. This is indeed the case for so-called generative AI, defined in the proposed AI regulation as systems « specifically designed to generate, with different levels of autonomy, content such as text, images, audio or complex video (generative AI) ». If the output has required an upstream reproduction of protected works, the benefit of the exception can be challenged because mining then contributes to the existence of a prejudice. The question of the threshold may be raised as the mining is not limited to a single work but is concerning large amounts of them.
Limitation of the TDM exception to cases where the result produced is not equipollent with a work or other protected object. Insofar as the three conditions of the test are cumulative, the absence of compliance with the first condition – the special case – for a sub-category of the cases covered by the TDM exception of article 4, namely the automated analysis of works to produce work-like information, is sufficient to question its compliance. Following this approach, the scope of the current mining exception should be limited to hypotheses in which the harvesting of works leads to the production of information that is not equipollent with works. Conversely, where the direct or indirect aim of the processing is to generate outputs that are equipollent with works or other protected objects, the exception would not be admissible, and the exclusive right would resurface. It would then be necessary either to negotiate agreements, or to draft a new dedicated exception shaped for the purpose of generating works, which could be accompanied by fair compensation, since it would then be possible to demonstrate that the generation of these work-like, by competing with the exploitation of protected objects, causes unjustified prejudice to the owners. In this way, one could remedy the free raid on protected content that feeds the automatic generation of quasi-works without compensation. This could be done without prejudice to any other mechanism that might deal with the output market[18].
II. AI-Generated Outputs Equipollent to Works Should Belong to a Domaine Public Payant
Musset used to say that a door should be open or closed. If, as a number of jurisprudential and doctrinal positions suggest, there can be no copyright on content generated by an AI system, in the absence of evidence of creative intervention by a human being, the result should be that such content is placed in the informational public domain, which would meant that the economic entities that are the source of the outputs of the machine cannot claim any exclusive right over this content. This approach is not only necessary to maintain the coherence of the category, it is also useful for those involved in creation, insofar as it rules out the possibility of using reservation tools to the benefit of entities which do not take the risks inherent in this economy, but which nevertheless compete with those who do. There are strong social justifications for ruling out plans for intellectual property substitutes for such objects, even if determining which objects fall within the public domain may prove tricky (2.1.). In any case, this entry into the public domain must not be accompanied by a systematic principle of free access, which would lead to unfair competition between AI outputs and intellectual property objects for which a royalty is payable. In order to prevent the free use of such content from being detrimental to protected objects, it would be advisable to introduce a remuneration system for these specific objects, in a Domaine public payant approach (2.2.).
2.1. The Public Domain Regime for Prima Facie AI-Generated Outputs
- The public domain, consequence of the non-protection of AI-generated outputs
No protection if not for humans – a disputed position. The ratio legis of intellectual property differs from one jurisdiction to another, but in principle it is based on a common meta-objective: to encourage individuals to create, so that society as a whole can benefit from this creation directly or indirectly. My starting point here is the idea[19], which has been the subject of some debate[20] that such an aim would be partially threatened if « authorless » productions were to be included within legal exclusivity regimes, where the monopoly would reward the sole investment in production or the use of so-called generative tools. On the contrary, some people consider it necessary to encourage investment in the development of AI by extending existing monopolies or adapting them to its characteristics[21] or they advocate the adoption of such reservation systems, while at the same time reducing their benefits through compensated exceptions. In this way, an intellectual property right would be recognized – there is still a lively debate as to the beneficiary between the technology provider and the user – but if need be, exploitation could give rise to some form of compensation for the prejudice suffered by the « human » authors affected by this unfair rivalry.
The reflex to move towards more intellectual property when the interest to be preserved is no longer that of human creators must be seriously questioned. On the contrary, by affirming the entry of such outputs into the public domain[22], the threat of a second-rate copyright can be removed, which, following the example of English law, offers protection to those who have merely used a machine to generate a work. Why, indeed, confer an exclusive right on entities that have generated such « products », given that patent law can already cover the inventiveness demonstrated by machine developers? This would be tantamount to rewarding the lack of enthusiasm for hiring and remunerating humans for their creative work[23]. On the contrary, in the absence of the justifications that buttress the application of intellectual property, the regime should be precisely the absence of protection and the competition of players.
The rejection of protection in the absence of human intervention in the current situation. Whether in the French tradition, which recognizes a protected work only in the presence of an individual creator, or more recently in the United States[24], which has just affirmed this principle, there is a broad consensus that a « quasi-work » produced mainly by generative AI should not benefit from copyright protection. The logical consequence of this assertion, though rarely stated as such, is that such productions should enter the public domain, which is the usual fate of objects not eligible for protection. Therefore, if a label, rather than using authors or performers, produces a new record made by an AI, anyone will be able to copy it and communicate it to the public without having to ask for authorization. There can be no legal exclusivity in this case. When a publisher thinks he can avoid hiring the services of a graphic designer and have his book cover produced by a machine, he retains the possibility of exploiting the output and illustrating his work in this way, but he will then have to bear the risk of someone else taking over this cover – subject to unfair competition – to make free use of it.
The virtues of public domain. This system rules out the use of legal exclusivity that reserve markets for exploitation, but the public domain is not synonymous with free final use. The novels of authors who have been dead for more than seventy years are not sold for free: publishers make a profit depending on the type of books they make available, but they do not enjoy the rents conferred by the intellectual property monopoly. Moreover, inclusion in the public domain is not a sign of society’s lack of interest in the products that appear in it, but rather a sign of society’s desire, with regard to these objects, to allow free competition, freedom of information and freedom of enterprise to prevail as the ferment of creativity, in place of the monopoly. Content in the public domain does not have to « wither away », and unprotection of these outputs is not an obstacle to investment in AI, since they can circulate freely between economic operators, without monopolies creating barriers to entry.
The absence of a specific public domain regime. The idea, singularly in France[25] lacks support. It will be recalled that in 2016, the draft law for a Digital Republic had set out to enshrine a legal regime for the informational public domain in a defunct article 8[26] that never saw the light of day due to strong hostility from the Ministry of Culture and the CSPLA[27]. Numerous theoretical works[28]demonstrate, however, the value of thinking positively about the category rather than as a relic, a default solution which, in the absence of a clearly defined regime, authorizes all forms of indirect reappropriation. The enshrinement of the public domain in copyright law therefore remains limited, even if it did make a timid but noticeable entry into the CSDM directive[29]. In fact, according to the cryptic wording of Article 14, « when the term of protection of a work of visual art has expired, any material resulting from an act of reproduction of that work is not subject to copyright or related rights, unless the material resulting from that act of reproduction is original, in the sense that it is the author’s own intellectual creation« . It is understandable that the European legislator wanted to prohibit the re-monopolization of works whose protection had expired, but that the risk of contradiction with the derivative works regime called for caution. The fact remains, however, that for the first time in EU law, the idea that a work in the public domain should not be (re)protected by copyright and related rights has been expressed, i.e. a prohibition on the monopolization of the object itself.
A regime aimed at excluding the creation of legal or contractual exclusivities. Proclaiming the principle that outputs essentially generated by artificial intelligence systems are usable by everyone and should not be subject to copyright or neighboring rights as such, is, in the absence of an ad hoc regime, the starting point for reflection to ensure that other forms of exclusivity are not indirectly constituted. For example, in some systems for generating images, music or other quasi-works, the general conditions of use imposed on those who operate the systems stipulate that the result produced « belongs » to the system designer, leaving to the prompter a simple user license. It seems to us that asserting that the AI-generated product belongs to the public domain would also help to settle this type of claim. If the prompter’s contribution is incidental, the output is in the public domain, and no one – neither the prompter nor the system’s suppliers – can claim ownership of it. If the human contribution is so tenuous as to bear no witness to any creative input whatsoever, there is no reason whatsoever to grant any copyright or neighboring rights to anyone. If, on the other hand, the contribution of a natural person is distinguishable and decisive, copyright could possibly be recognized for the benefit of this natural person who use the AI system, by virtue of the intellectual control he maintains over the creative process, if the condition of originality is met.
- The subtle take of defining the perimeter of the elements of the informational public domain
Outputs that would be equipollent with works or other protected objects. While the logic of the public domain has its advantages, the establishment of a regime only makes sense if we can clearly identify the elements that fall within its scope. This is no minor difficulty when it comes to AI-generated content. The first question concerns the nature of these informational productions. Indeed, these systems are capable of generating all kinds of content, of producing all kinds of actions – driving vehicles, performing surgery, etc. – and the lightning development of the technique suggests that we are only at the beginning of the phenomenon. However, it is not a question here of reasoning about all the results produced thanks to or with the help of or by artificial intelligence, but rather, as we said earlier, of focusing specifically on the intervention of artificial intelligence systems in the field of intellectual creation, which presents a high risk of destabilizing the ecosystem. The criterion of decisive human intervention makes it possible to distinguish between what necessarily enter within the public domain and what might escape it.
- The intervention in the field of intellectual creation proves to be a rather unclear criterion
The absence of borders of the public domain. If we consider the public domain as the negative of intellectual property, the assertion that AI productions are excluded from protection should lead us to make a “decal” of the objects included in the informational public domain on the scope of intellectual property. This is the model on which the public domain operates for works whose protection has lapsed, since only objects that fell within the scope of intellectual property rights were covered. This would include quasi-works, pseudo inventions or designs. But such a transposition is impossible insofar as, on the one hand, the scope of application of intellectual property is fluid, so that it is not determined once and for all, and because, on the other hand, the material scope of application of the public domain is, in any case, not stackable on that of intellectual property, since it includes productions which, precisely, do not fall within its scope.
Copyright now embraces works that those who wrote the first copyright laws had no idea that they would come into being in the future. We speak of « literary and artistic property », but the vocabulary has since long been twisted to absorb creations that have only distant links with the original meaning of this expression: software, databases, bolts, salad baskets… The difficulty is even greater for patents, which do not positively determine the field of intervention: the invention can nestle in any sector of activity. While protection is sometimes refused to certain types of invention in specific fields (life sciences, etc.), these exclusions do nothing to help define the identical contours of the public domain, since this category precisely absorbs, by default, everything that cannot be protected. In this respect, the expression that ideas are free, a genuine emblem of the public domain, or the absence of protection of « raw » information, hardly seal the boundary between what would certainly fall within the public domain and what would be eligible for protection, since ideas and information are also included in protected objects.
The composite nature of the « informational » public domain. The superposition of scopes of application is therefore not possible: the material scope of the public domain is broader than that of intellectual property, and the productions generated by artificial intelligence systems would consequently fall within a composite set of elements, some of which have been protected by intellectual property, others of which are expressly excluded from it, or even fall outside its scope altogether. The public domain is a vast category, capable of encompassing the productions of generative AIs, without this leading to any clear identification of the items concerned. This lack of differentiation is not a problem at this stage, but it will have to be resolved in the event of the implementation of a domaine public payantthe scope of which would then be less vast than the public domain in general and would be homothetic with that of objects likely to be protected by intellectual property, as we shall consider later.
- The need for a criterion of (non)decisiveness of the intervention of a natural person
- Need of the criterion
Degrees of human /machine intervention. While it is not easy to determine with precision the nature of the results produced by AI that enter the public domain as a composite whole, it is possible to argue that the regime should apply in the absence of decisive human intervention in the result produced[30]. Here again, the boundaries are porous, and the delimitation exercise is perilous[31]. Artificial intelligence, so misnamed, is not yet fully autonomous, and the implementation of systems still requires human intervention. Individuals supervise the machine’s actions, others program and improve them, others pilot them to achieve the desired results, and so on. In this way, the human element is still present at all stages of the development of an output produced by generative AI. I don’t say that any creation resulting from the use of generative AI systems by a human being should be excluded from copyright, neighbouring rights, design rights or patent rights, but rather that a simple implementation of the machine’s functionalities or a financial investment in the creation, purchase or manipulation of the tool should not give rise to monopolization of the result produced.
Criterion of « intellectual » decisive human intervention in copyright. yes. The criterion of decisive human intervention seems to be emerging[32] in the still incipient case law and doctrine[33]. The personalist conception of French law is in line with such a criterion insofar as it accepts no author other than a natural person, even if, at the margin, the mechanisms of ownership allow a legal entity to be treated as an author[34]. In other words, the foundations of copyright presuppose an intellectual human intervention whose intensity is such that it extends to a condition of originality requiring the imprint of one’s personality in the work produced. Even when watered down by European law[35], the criterion of an intellectual creation specific to its author still seems to require the intervention of the brain (intellectual creation), which allows a form of identification of its contribution (specific to its author), in other words that the intervention of the human-author be identifiable as such in the work[36]. Other countries, such as the USA[37] and Australia[38], have already used this criterion of absence of human intervention to deny copyright protection to AI-generated outputs, or to conclude that protection should be applied in a distributive way according to the degree of intervention of the author and the machine. Without going into detail about the fundamental principles underpinning copyright, the point is to assert its relevance as a criterion for selecting what should be protected and what should not.
Criterion of decisive bodily human intervention for performers’ performances. The question of neighbouring rights is more delicate, as the thresholds imposed to qualify for protection are less clear and the requirement of human intervention less obvious. Performers’ rights are without doubt the least problematic in this respect. Here, human intervention is of a different nature, since it is more corporeal than intellectual, the artist being the human intermediary through whom the work will be perceptible to the senses of an audience, itself made up of human beings. Moreover, the neighbouring right is not recognized for the so-called « complementary artist », insofar as the latter’s intervention is not decisive in the performance. The requirement for human intervention here is twofold, if we consider that the artist’s neighbouring right can only apply to a performance provided by a natural person on the basis of a work itself created by a natural person. If the human being is no longer involved in the interpretation of his or her performance, it is undesirable for a monopoly to be granted to the operator who will trigger a machine-generated interpretation, or to the supplier of the technology, a fortiori if the latter borrows features from existing artists.
Decisive intervention in the granting of investors’ neighbouring rights: a criterion to be imposed. The difficulty lies more in the granting of neighboring rights linked to investment efforts in content production. Here, there is no trace that decisive human intervention criterion can be used to distinguish between what can be protected and what must not: the beneficiaries are essentially legal entities, and the monopoly is granted to them in return for the economic risk taken in financing the production or distribution of “cultural” content. What’s more, the notion has evolved in such a way that rights are recognized even where no protected work is involved: the program for the broadcaster, non-original databases for the database producer, and so on. Courts have even recognized the absence of the need for a prior work for the producer of phonograms, vesting a right in the mere fixation[39] of a birdsong[40]. In this perspective, how the criterion of decisive human intervention can serve to circumvent the possibility of monopolizing AI-generated content, to the benefit of those who have invested in the use of such systems, when this criterion is not required for other types of production?
Several arguments could be put forward in this perspective. Firstly, the notion of neighbouring right could be revived, as could the principle of articulation between copyright and neighbouring rights, which presupposes that the latter do not impede the former[41]. In this hierarchical perspective, it is difficult to see why a monopoly should be granted to the producer of a melody generated entirely by an AI, on the grounds of mere sound fixation, whereas no copyright could be recognized in this case. Furthermore, the phonogram producer’s right would persist if the AI-generated melody is interpreted by a performer, whereas in the absence of a protectable work, it is doubtful that the performer would be vested in any neighbouring right. The destabilization of relations between owners, linked to the absence of a requirement for decisive human intervention, augurs serious difficulties, for example with regard to equitable remuneration for phonogram broadcasting. Indeed, if it is sufficient to fix a recording of a song produced exclusively by an AI in order to enjoy an exclusive right, there is a risk that the investments[42] made by producers in the production of works created by human authors and performed by human artists will drop drastically in favor of less demanding and possibly more profitable AI-driven projects. There’s no reason to grant a monopoly in this case, and the public domain regime, enabling everyone to reproduce and use the objects produced freely, would be a good way of continuing to encourage producers to invest in human creation.
Finally, while the criterion of decisive human intervention is not expressly found in the texts, it can nevertheless be inferred from the ratio legis. Not only were the rights of producers of phonograms and videograms initially granted to encourage the production of works of the mind, and thus creations involving human intervention, but the same is true for press publishers, whose claim to an exclusive right over press publications, which would be generated by AI and not by human journalists is hard to envisage. The mention of journalists’ rights in the mechanism of the press publisher’s neighbouring right, and their vocation to benefit indirectly from this right, shows that the legislator did not envisage this mechanism without the intervention of human-author journalists. The question is more delicate for database producers, who could, if necessary, continue to claim their sui generis right on AI-generated databases, provided they can prove that the investment made is quantitatively or qualitatively substantial. The significant reduction in costs associated with the use of such systems could justify the refusal of protection, on the grounds that the investment would not be sufficiently substantial.
The absence of decisive human intervention as a dividing line between reservation mechanisms and the public domain. Ultimately, it seems that adopting the principle that any AI-generated output that does not involve decisive human intervention in the creative process should fall within the scope of free competition and the public domain would be opportune both to maintain the balance between rightholders and to stimulate investment in the production of human creations. However, the concrete elements needed to implement this criterion are still lacking.
- The evidence of the decisive human intervention and the possibility of a presumption
Verification of human/machine intervention. Adopting a criterion to distinguish between what is in the public domain and what is not is of no use if it is excessively hard to determine. The evidence of decisive human intervention could prove difficult to provide, given that the creation process is opaque, and that stakeholders may have every interest in remaining evasive on the question if they fear losing the benefit of an exclusive right by revealing the degree of intervention of the machine. However, this objection can be overcome. This is not a new issue in copyright law, where untimely claims and fraud are commonplace, nor is it new to the discussion of transparency that accompanied the adoption of the regulation on Artificial Intelligence. This transparency requirement is now enshrined in the text and is constantly being reinforced. Individuals need to know whether the services they receive or the products they use have been created using artificial intelligence. It’s still important to be informed about the quality of the interlocutor: am I talking to a human or a chatbot? Has the decision been taken by human beings, or is it simply the product of an algorithm? The examples are numerous, and the value of this transparency obligation, which goes far beyond questions of copyright, is not in doubt.
Transparency and its limits. Those who use generative AI systems will have to disclose this fact, so that the recipients of the objects produced by AI are informed. This transparency obligation will certainly make it easier to prove, in the negative, that human intervention was decisive, but it will not be enough. As already mentioned, many hybrid objects will be produced using such systems, without this necessarily disqualifying all intellectual property rights, the output not necessarily being included in the public domain. So invoking transparency cannot be the key to resolving all difficulties. There will undoubtedly be an interest in gradations in the degree to which AI is used to refine the model. A measure of the intensity with which the AI system is used would approach the proof of the proposed criterion by approximation. If the indicator specifies that the object is produced exclusively by an automated mechanism, the criterion of decisive human intervention will certainly be lacking, but there will undoubtedly be many in-between situations. Identification techniques could perhaps provide a clearer idea of the respective proportion of AI system use and human intervention in the future, but all this is still under construction, so it is difficult to rely exclusively on this type of solution.
Presumption of non-intervention due to the non-attribution of authorship to a natural person. It is therefore appropriate to add another evidential mechanism, without however placing an impossible burden of proof on rightholders: that of not having used an AI system. The prohibition of formalities in copyright law prevents the benefit of protection from being subject to such a demonstration. There will certainly be a debate as to the reality of the author’s intervention in an infringement dispute, but there cannot have any precondition of non-declaration of use of an artificial intelligence tool to benefit from the right. However, it is important to determine which elements fall within the scope of the public domain, and in particular which of them should generate payment of remuneration in the event of use. The mechanism of presumption can play a role. When a work is disclosed under a person’s name, that person is presumed to be the author, in the absence of proof to the contrary, under article L. 113-1 of the French Intellectual Property Code[43].
It would be sufficient to invert the proposal and state that if AI is used, which in principle will be subject to a disclosure obligation, and no attribution is made to a natural person designated as author, the output would be deemed to be in the public domain. If the exploitation of the output does not give rise to any attribution to at least one person designated as the author, likely to claim a moral right as well as an economic right, it would be presumed to be in the public domain. The presumption is, of course, a simple one, so that the absence of credit for a human being when an AI is used will not prevent an author from claiming rights if he manages to show the decisive character of his intervention and its originality. In order to avoid false attribution, it is conceivable, following the example of American law, that a claim to a literary and artistic property right made when no human person has been decisively involved in the process of creating the output could give rise to a sanction.
Assessing decisiveness. It remains to be seen what will be decisive in the human intervention. This question relates more generally to the criteria for determining the status of author, performer, or even that of other owners, and the exclusions that surround them. Is the intervention solely intellectual, or is it also manual? What is the weight of each author, when the work is created by several people, to reach this threshold? Has the human being actually played a role in the conception, or is it confined to execution tasks, etc.? In reality, these questions are quite common, and they will have to be clarified by case law.
Assuming that all discussions concerning proof of public domain status have been settled, when the output is in the public domain and its use is “competitive” with/ detrimental to that of human production, a compensatory mechanism should be triggered, to rebalance conditions in the markets concerned. Several systems are conceivable: a higher VAT rate, a tax allotted to certain bodies, but there is one that has its roots in the history of intellectual property, and which seems more appropriate for organizing a virtuous system for human creators: the domaine public payant.
2.2. Model(s) of Domaine Public Payant
Free use for a fee. The domaine public payant is based on a simple idea: to avoid “competition” between living and dead authors. This mechanism has been the subject of intense legal writings and has recurred in major copyright debates in France and around the world. It has met with varying degrees of success and has manifested itself in many ways. It is not unrelated to the situation we are dealing with here, namely the “competition” between human authors and machines[44].The premise is that the outputs generated prima facie by machines are not doomed to be protected by intellectual property rights, as there is no social interest in rewarding with a monopoly those operators who prefer to use these tools rather than hire human creators. Their destiny is therefore to be in the public domain and freely reusable, within the framework of open competition. But it would be counterproductive for the encouragement of human creativity if their use were free of charge.
Of course, an operator using these ersatz products would run the risk of being copied without being able to oppose the use of infringement tools, but he could always activate the threat of unfair competition or parasitism to oppose certain reproductions if the conditions for these actions were met. We therefore need to add a second deterrent tool namely, to make the use of these outputs less attractive by introducing a system of domaine public payant. While this mechanism has several advantages, its implementation requires further clarification to shape the tool for the hypothesis at hand.
- The idea of domaine public payant from Victor Hugo to Adolf Dietz
Variety of models in time and space. There is not one, but several models of the domaine public payant. The idea originated in the 19th century, notably with Alfred de Vigny[45], Pierre-Jules Hetzel[46] and, more spectacularly, Victor Hugo[47], who passionately defended it during the speeches that led to the creation of ALAI. Originally, it was closely linked to extending the term of post-mortem copyright protection. Several variants were proposed: deducting part of the royalties from works used after the author’s death, taxing operators who preferred to use unprotected repertoire authors to the detriment of living authors, and so on. The notion of the public domain has expanded and been applied in a variety of ways in different countries.
In France, this is illustrated by the system of remuneration levied on unprotected works due to a lack of reciprocity, and allocated for cultural actions, although the system has never really been enforced[48], or by the conclusion of contracts with theater operators to mutualize tariffs for the representation of playwrights, in order to encourage them to perform the contemporary repertoire, etc. Some countries[49], like Argentina[50] have implemented a real tax levy on any use of a work in the public domain, including derivative works[51].Thus, from time to time, the mechanism has gone beyond the sole hypothesis of works whose term of protection has expired, to cover other situations. It has borrowed from several levy systems – tax, private law – and led to various forms of redistribution – to the benefit of the State, cultural action, maintenance of the public domain -. While a number of countries have adopted such mechanisms[52], which are mentioned notably in the Bangui agreements[53] they are still somewhat residual and limited in scope[54]. Nonetheless, the notion of domaine public payant has always been a formidable laboratory of fertile ideas[55].
Several proposals include the mechanism within the ambit of the legislation dedicated to copyright. The Jean Zay bill on droit d’auteur, of 13 August 1936 encompassed an article 21 where the last forty years of the post-mortem fifty years protection were subject to a kind of right to remuneration allowing to publish ou reproduce the work against the payment of a fee. This provision that had been highly debated was nevertheless scheduled for discussion before the Parliament in Novembre 1939. Only the war changed the plans and the whole bill was definitively buried.
Collectivist approach to the domaine public payant. As early as 1976, Adolf Dietz, a German scholar, also promoted the idea that the paying public domain was a matter not of taxation but of copyright law, proposing that a new subject – the community of authors – should be the beneficiary of a share of the profits generated by the exploitation of public domain works[56].This collectivist approach is reminiscent of work aimed at associating the actors of traditional knowledge and folklore with the commercial exploitation by third parties of objects derived from them[57][. In this perspective, the protection of these elements by intellectual property rights would not be totally suppressed, and communities could receive some form of compensation in counterpart of unauthorized extraction.
Le domaine public payant as a tool within the Intellectual Property regime. Adolf Dietz’s idea of a community of authors’ rights was to set aside the possible tax dimension and focus on a form of continuity and solidarity between generations of authors. However, it was based solely on the perspective of a contribution levied on the exploitation of works that were no longer protected after the end of the legal period of protection, and not on a broader vision of the public domain. The interest of the present approach is, as mentioned above, to integrate the regime of the public domain in general and the domaine public payant in particular within the intellectual property system, so that it constitutes an instrument of internal equilibrium and not a totally external mechanism, which would come under the sole common law of the notion of common things. Attached to intellectual property law, the tools of the informational public domain and its “for remuneration” sub-declination would be legitimately exercised by private entities, for various purposes such as the promotion of culture and innovation by humans.
- The implementation of the mechanism
Sandbox. The present article is not intended to resolve all the implementation difficulties that the introduction of such a system raises. There are many, and not all of them can be anticipated. Rather, while presenting the idea, it is an attempt to measure its feasibility, without necessarily determining all the choices involved in applying the system. This second « sandbox » phase presupposes that one already subscribe to the very principle of the mechanism, and the present analysis is, for the time being, no more than a bottle thrown into the sea in this direction. Some of the constraints and open options will be considered in turn.
- The constraints
Legislative intervention is both necessary and possible. The first constraint is that setting up a Domaine public payant requires legislative intervention. At present, the status of AI-generated outputs is unclear, and the market is moving either towards free use of everything that, to a greater or lesser extent, required the intervention of AI systems in its production, or towards questionable claims to intellectual property on the part of the suppliers of the systems in their UGCs. Clarification is therefore required by law, on two levels: on the one hand, to assert that such content belongs in the public domain, and that it is therefore impossible to claim intellectual property rights over it; and on the other, to introduce a compensatory mechanism in return for the use of these outputs.
On the other hand, if the French legislator opts for a Domaine public payant, it is not bound by the acquis communautaire, insofar as this issue has not been harmonized, with the exception of article 14 of the aforementioned CSDM directive. Consequently, it is fully competent to intervene with regard to the exploitation of these outputs on the French territory. The difficulty undoubtedly arises from the risk of « pulling the trigger » first and introducing such a mechanism when the use of the same objects remains free of charge in other countries. We can no doubt anticipate discontent on the part of remuneration payers, who will not fail to point out that copyright holders are preventing innovation by making access to these productions more expensive through a disguised form of taxation, and so on. So be it. But one must start somewhere. Besides, it’s not out of the question for other countries to follow suit, as the issue is becoming so acute in countries with a strong creative economy. Above all, in case the payment would be linked to the exploitation of these outputs, the link to the national territory would therefore avoid the risk of circumvention towards a lower-cost regime. In addition, as mentioned above, the current solution, which allows free riding to develop, with the possibility of recreating exclusivities for entities that are already in a position of economic, technological and contractual power on the market, is unsatisfactory.
Impact analysis. If the aim of the mechanism is to dissuade operators from turning to AI-generated outputs, yet it must not be excessively complex and penalize the development of AI. This should not be the case if the mechanism clearly defines its scope of intervention, and if remuneration is paid only on the production or exploitation of outputs that are equipollent with protected works or objects. However, a detailed economic analysis is a prerequisite for setting up a paying public domain namely, to determine the appropriate levy levels in line with the public policy objectives pursued.
- The pending options
Sky is the Limit. The variety of existing models for Domaine Public Payant, as well as their relative lack of effectiveness, allows the imagination to run wild and, at this stage, enables us to envisage a variety of options on the questions such as the event giving rise to remuneration, the debtors, the amount of remuneration, its collection and its allocation.
Generating event. When it comes to the triggering event for remuneration, there are essentially two options: the exploitation[58] of an output generated by an AI, or the simple use of a certain technology. The first hypothesis is undoubtedly the most obvious, insofar as it is based on an event occurring in the same market as that in which the intellectual property owners find themselves, i.e. the act of exploiting substitutable objects. The question then inevitably arises as to the type of use subject to the law – commercial exploitation, private use, use of the first-level output, incorporation of all or part of the output into a complex work…
The second is upstream of this market, and captures the production process itself, irrespective of the success of the exploitation of the output; in this case, it would be the use of artificial intelligence systems to generate this type of output that would be concerned. However, both proposals presuppose not only that the use of the technique is identified or identifiable in the output, but also that the type of technology targeted is circumscribed by the text, with the inevitable risk of obsolescence of the mechanism if this definition is too restrictive. The technical approach also allows for possible modulations, such as considering the negative externalities of using an energy-intensive technology (the cost of the prompt), or a form of guarantee of the lawfulness by design of the proposed solutions (traceability of the content fed into the training data, protection of personal data). The exercise is a difficult one, but it could accompany the introduction of regulations on artificial intelligence and provide an opportunity to encourage operators to provide more « responsible » technical solutions.
Possible debtors of the remuneration. This raises the question of who should be liable for payment of the remuneration, which, in the first instance, suggests targeting the persons exploiting these outputs, insofar as they are the beneficiaries of the efficiency gains linked to the use of productions that are less costly than human productions as regards the remuneration of creation. It would therefore be logical for the burden to fall on them. The problem of transparency and possible fraud must be anticipated as far as possible. How can we know that the operator is using an object produced with artificial intelligence? Wouldn’t it be possible to circumvent the system by using vague attribution tools, like DRs for photographic credits? Wouldn’t it be more appropriate to collect the “tithe” from the companies that supply the technology, insofar as the traceability of their intervention is possible? They also benefit from their customers’ enthusiasm for the productions generated by their tools, while making a profit when the use of these tools gives rise to royalties. Should the user of the technology (the one who triggers the output) be involved, or should the debtors all be involved? Would it be advisable to distinguish, as is the case for private copy levies, between those who cause the damage (the user) and those who actually owe the compensation (the manufacturers), so as to build efficient economic models for collection by centralizing debtors?
Amount of remuneration. Remuneration may be fixed on a flat-rate or proportional basis, and determined according to a certain base and rate. For example, article 68 paragraph 2 of the Bangui agreements stipulates that the amount of remuneration for the exploitation of works in the public domain must be equivalent to half of the remuneration normally paid to the owners of copyright and related rights in their protected works and productions, in accordance with the contracts or practices in force. Would this be enough to dissuade operators from turning to AI-generated content in favor of human creations? Should we draw inspiration from existing compensatory mechanisms such as remuneration for private copying or the equitable remuneration system for broadcasting commercial phonograms, and consider setting royalties at the end of a negotiation process between market players, or even by administrative commissions determining mandatory tariffs? If modulation of the mechanism according to the technology used is an option chosen, it could also be reflected in the level of remuneration rates, etc. Here again, the options are legion, with their respective advantages and disadvantages to be weighed up by those responsible for implementing the public policy for which the instrument of the public domain can be fine-tuned.
Collection and redistribution. Bearing this in mind, it remains to consider the mechanisms for collecting and redistributing the sums collected. Various models exist, from mandatory collective management to collection by an ad hoc body, and others can be created depending on the final destination of the remuneration collected. Direct redistribution to rights holders seems difficult to achieve insofar as – apart from outputs whose similarity to an existing work could assimilate it to a copy – it will undoubtedly be extremely difficult to establish an individual link between the prejudice suffered by one specific author and the exploitation of AI-generated content[59]. The paying public domain is similar to a global mechanism, closer to a support fund than to an insurance system that would distribute to each individual the amount of the prejudice suffered. It seems to me, therefore, that the system should be one of allocating sums for collective purposes, enabling an indirect redistribution of the wealth produced to the community of living authors, since the aim is to promote human creation. Here again, existing mechanisms may provide inspiration for a realistic solution: the allocation of 25% of private copy or of royalties that could not be distributed for the promotion of cultural action, the use of certain sums by OGCs for social assistance to their members, etc. The definition of redistribution mechanisms is, once again, a function of public policy objectives: aid for young creators, social assistance for the poors, training artists in new tools to support the digital transition, maintaining alternative technical know-how in automatic production to support energy sobriety, organization of subsidized events…
Conclusion. To end with, the sheer number of possible solutions makes one’s head spin, providing a brighter horizon than the bitter reality of a technical advance that may be working against humanity. Generally speaking, we must be wary of thinking that the apparent novelty of a phenomenon necessarily engenders the novelty of the legal response that accompanies it. Before embarking on novel regimes whose effectiveness has yet to be proven, and whose legal complexity is high (such as neighbouring rights for robots, or copyright for AI system providers), it is possible to look to existing concepts for the seeds of a possible solution. Thus, a bold reading of the three-step test could partially disqualify TDM’s free exception. Furthermore, there is an old idea that may not solve all the problems, but could be useful to the copyright economy, which sees the onslaught of AI as an existential threat: it’s called the paying public domain.
There is no providential solution to the transformation that is about to take place and which will affect the whole of society. New things will emerge, while others will obviously be destroyed. But a combination of mechanisms, ranging from the avoidance of informational asymmetries through transparency obligations, to the consideration of AI’s negative externalities to encourage the emergence of a responsible technology, via training in the tools and/or subsidy logics, combined with the compensatory systems studied, will perhaps make the industrial revolution that is underway less painful for the community of human creators.
[1] This article shares many ideas with the paper recently published by M. Senftleben, Generative AI and Author Remuneration. IIC 54, 1535–1560 (2023). https://doi.org/10.1007/s40319-023-01399-4 Available at SSRN: https://ssrn.com/abstract=4478370. If the subjects are commons, the opinions are sometimes diverging as underlined below. The present paper is a work-in-progress and shall benefit from your comments. I could not quote all the relevant literature.
[2] T. di Lampedusa, the famous sentence pronounced by Don Fabrizio Corbera, prince of Salina in the novel Il Gattopardo.
[3] This concept is still very approximatively defined. See article 28 ter in the version of the Parliament (the position of the trilogue is still pending in the AI Act project at the time this paper is published).
[4] The first week of exploitation the song raised the first position of the Digital Song Sales chart de Billboard, which is based on the number of downloads in the US.
[5] Notably on the consequences of AI on labour, see the impressive review of social sciences articles, Deranty, JP., Corbin, T. Artificial Intelligence and Work: a Critical Review of Recent Research from the Social Sciences. AI & Soc (2022). https://doi.org/10.1007/s00146-022-01496-x
[6] This option is, among others, stressed by the Report to the CSPLA on AI and Culture A. Bensamoun, J. Farchy, P.-F.Schira, rapporteur, on the outputs generated by AI: “In this respect, it is important to note that copyright seems flexible enough to include these creations and that the attribution of rights to the AI creator appears likely to provide relevant solutions. This is what a Chinese court seems to have decided, to the benefit of the Tencent company. As such, unless jurisprudence highlights a new need or refuses to receive these creations, positive law should, for the time being, be able to be applied, through a new interpretation of the criteria for accessing protection.” p. 46.
[7] J.-M. Deltorn, Droit d’auteur et créations des algorithmes d’apprentissage, PI janvier 2016, n° 58, p. 4.
[8] Face au développement des outils d’intelligence artificielle (IA), la SACEM annonce exercer son droit d’opposition (opt-out) au profit de ses membres, Légipresse 2023. 519 — 27 octobre 2023.
[9] M.R.F. Senftleben (2004), Copyright, Limitations and the Three-Step Test: An Analysis of the Three-Step Test in International and EC Copyright Law, The Hague/London/New York: Kluwer Law International 2004, according to whom the opt out faculty prevent the existence of the harm caused to normal exploitation; see M.R.F. Senftleben (2014), “How to Overcome the Normal Exploitation Obstacle: Opt-Out Formalities, Embargo Periods, and the International Three-Step Test”, Berkeley Technology Law Journal Commentaries 1, No. 1 (2014), 1 and lately Senftleben, Martin, Generative AI and Author Remuneration already quoted.
[10] R. Ducato, A. Strowel: Ensuring Text and Data Mining: Remaining Issues with the EU Copyright Exceptions and Possible Ways Out, European Intellectual Property Review 43 (2021), 322.
[11] A. Bensamoun, To be or not to be…transparent – Pour un principe matriciel de transparence dans l’environnement numérique, Dalloz Actualité, 6 décembre 2023.
[12] In the article L. 112-2 of the French Code of Intellectual Property, there is no definition of the work but a presumption of what a work of the mind is by an unlimitefd list of works supposed to be so in various creation fields such as film making, books, photography, drawings…
[13] CJUE, 11 juin 2020, Brompton Bicycle, C‑833/18, EU:C:2020:461, para 22 to 25.
[14] On these already old questions, E. Ulmer, Les problèmes de droit d’auteur découlant de l’utilisation d’œuvres protégées par le droit d’auteur dans les systèmes automatiques d’information et de documentation, Le DA, février 1978, p. 67. « Dans le cas de la méthode des index, la mémorisation et la récupération des données bibliographiques (auteur du texte, titre, éditeur, etc.) ne constituent généralement pas une atteinte au droit d’auteur sur les œuvres auxquelles se rapportent ces données » ; Le DA, n° 7-8 juillet-août 179, p. 196 « Le groupe de travail a été d’avis qu’il n’y avait pas d’atteinte lorsque les indications usuelles de l’auteur, du titre, de l’éditeur, etc (méthode des index) sont mises en mémoire sur ordinateur. » ; F. Gotzen, Le droit d’auteur face à l’ordinateur, Le DA janvier 1977, p. 19 « Restera libre la mémorisation d’index ou de tables qui ne contiennent que des données bibliographiques ou des mots-clés, car il n’y a pas alors reproduction des œuvres mêmes. »
[15] C. Cass 1. civ., 9 novembre 1983, n° de pourvoi 82-10005, and C. Cass. Ass. Plénière, 30 octobre 1987, n° de pourvoi 86-11918.
[16] CA Paris, 26 janvier 2011, SAIF c/ Google.
[17] CJUE, 5 juin 2014, no C-360/13, V.-L. Benabou et F. Gaullier L’arrêt Meltwater de la CJUE, fonte ou refonte des droits d’auteur ?, Légipresse, oct. 2014, p. 539.
[18] See contra M. Sentfleben, quoted. note 1.: “To achieve this goal, it could be clarified in the context of AI levy legislation that individual authors and right holders can only benefit from levy payments and related repartitioning schemes and social and cultural funds of collecting societies when they refrain from invoking Article 4(3) CDSMD to reserve their rights.”, p. 17.
[19] Sharing this human-centered approach encompassing also patents, F. Pollaud-Dulian, L’humanisme de la propriété intellectuelle au défi des objets produits par intelligence artificielle, D. 2022.2051.
[20] See for example, the French Report for the CSPLA on Artificial Intelligence, A. Bensamoun, J. Farchy, sur ce questionnement. D. Gervais, « Can Machines be Authors », http://copyrightblog.kluweriplaw.com/2019/05/21/can-machines-be-authors , May 2019 ; La machine en tant qu’auteur, Propr. intell. 2019, n° 72, p. 7 ; P.-Y. Gautier, De la propriété des créations issues de l’intelligence artificielle, Revue pratique de la prospective et de l’innovation, n° 2, octobre 2018, Dossier 12.
[21] I. Randrianirina, Plaidoyer pour un nouveau droit de propriété intellectuelle sur les productions générées par intelligence artificielle, D. 2021. 91. See also N. Enser, Conscience de la création et droit d’auteur, coll. CEIPI, LexisNexis, pleading for a need to rethink the protection, p. 268 and sq.
[22] G. Azzaria, IA et création non appropriable, https://www.lesconferences.ca/videos/mise-en- commun-numerique et Intelligence artificielle et droit d’auteur: l’hypothèse d’un domaine public par défaut, Les Cahiers de PI, 2018, vol. 30, n° 3, p. 925.
[23] See for a similar position, Picht, P.G., Thouvenin, F. AI and IP: Theory to Policy and Back Again – Policy and Research Recommendations at the Intersection of Artificial Intelligence and Intellectual Property. IIC 54, 916–940 (2023). https://doi.org/10.1007/s40319-023-01344-5 “However, once an AI system has been developed, it can produce content such as text, images, music, films, and the like at almost zero marginal cost. While it may be important to grant some form of IP protection for the AI system, there is no need to incentivize the use of these systems by granting copyright protection to their output”, p. 923. See also European Commission, Directorate-General for Communications Networks, Content and Technology, Study on copyright and new technologies – Copyright data management and artificial intelligence, Publications Office of the European Union, 2022, https://data.europa.eu/doi/10.2759/570559, p. 232 and sq.
[24] See the recent decisions of the Copyright Office and the Guidelines published in March 2023:Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, Federal Register, vol. 88, n° 5188, 16 mars 2023). ; US District court for the District of Columbia, 18 août 2023, Stephen Thaler c/ Shira Perlmutter, n° 22-1564, BAH ; Copyright Review Board, 5 septembre 2023, sur la demande d’enregistrement de l’image Théâtre D’opéra Spatial, générée à l’aide du logiciel Midjourney. N. Enser, IA générative : le début des difficultés, ou quand l’IA et l’humain concourent à la création, D. actu. 13 oct. 2023 ; N. Enser, L’« Entrée dans le Paradis » du droit d’auteur : pas sans un être humain à l’origine de la création ! Dalloz actualité, 18 sept. 2023.
[25] See the unavoidable study made S. Dusollier for WIPO ; https://www.wipo.int/export/sites/www/ip-development/fr/agenda/docs/scoping_study_cr.pdf
[26] L. Maurel. La reconnaissance du “ domaine commun informationnel ” : tirer les enseignements d’un échec législatif. Vers une République des biens communs?, Les Liens qui Libèrent, 2018. hal- 01877448.
[27] Despite of the moderated conclusions of the report of J. Martin for CSPLA, Mission du CSPLA sur le domaine commun informationnel. 12 October 2015: http://www.culturecommunication.gouv.fr/Thematiques/Propriete-litteraire-et- artistique/Conseil-superieur-de-la-propriete-litteraire-et- artistique/Travaux/Missions/Mission-du-CSPLA-sur-le-domaine-commun- informationnel.
[28] The comments are numerous in the Anglo-Saxon doctrine. For French-speaking contributions, among others. S. Choisy, Le domaine public en droit d’auteur, IRPI-Litec, 2001, p. 226, n° 464 ; M. Clément-Fontaine, L’œuvre libre, Larcier, 2014 ; M.-A. Chardeaux, Les choses communes, LGDJ, 2004 ; S. Vanuxem, Les choses saisies par la propriété.De la chose-objet aux choses-milieux, Revue interdisciplinaire d’études juridiques 2010/1 (Volume 64),p. 123 ; S. Dusollier, Du commun de l’intelligence artificielle, in Penser le droit de la pensée, Dalloz, p. 107, 2020 ; Pour un régime positif du domaine public. Benjamin Coriat. Le retour des communs, Editions les Liens qui Libèrent, 2015 ; V.-L. Benabou et S. Dusollier, Draw Me a Public Domain, in Coyright Law, Handbook ed. P. Torremans, Elgar, 2007, chapter 7.
[29] Namely recital 3 quoting recital 53.
[30] For this position the Report of the French group of AIPPI (Association Internationale pour la Protection de la Propriété Intellectuelle, 2019 – Question Study – Copyright in artificially generated works : https://aippi.org/wp- content/uploads/2019/06/2019_FR_2019_Study_Question_Copyright_in_artificially_generated_works_2019-05- 24.pdf. ; J. Cabay, Contre la protection du droit d’auteur pour les productions de l’intelligence artificielle, nov. 2018, https://orbi.uliege.be/handle/2268/232795;http://hdl.handle.net/2268/232795 .
[31] Senftleben, M. and Buijtelaar, L., Robot Creativity: An Incentive-Based Neighboring Rights Approach (October 1, 2020). Available at SSRN: https://ssrn.com/abstract=3707741, which suggest the possibility albeit not an easy one to create a neighbouring right, which would be a right to equitable remuneration for machine-generated «content » for the benefir of the user of the machine. See. R. Yu, The Machine Author: What Kind of Copyright Level Protection is Appropriate for Fully Independent Computer-generated Works?, University of Pennsylvania Law Review 165 (2017) 1245 (1245-1246) who considers that copyright protection is not suitable. M. Perry & T. Margoni, “From Music Tracks to Google Maps: Who Owns Computer-generated Works”, Computer Law and Security Review 26 (2010)?
[32] The de minimis of criteria for the intervention of the machine seems relevant for the Copyright Office to allow or not copyright protection.
[33] J. C. Ginsburg et L. A. Budiardjo, Authors and machine, Columbia Public Law Research Paper No. 14- 597 ; Berkeley Technology Law Journal, Vol. 34, No. 2, 2019. https://ssrn.com/abstract=3233885 or http://dx.doi.org/10.2139/ssrn.3233885, considering that to be author, the person must have the control of the conception and of the execution of the output. See also N. Enser, Conscience de la création et droit d’auteur, CEIPI, LexisNexis, 2022.
[34] The French Court of Cassation said that a legal person cannot have authorship: Cass. 1ere civ., 15 janv. 2015, n° 13-23.566, D. 2015, p. 206 et p. 2215, obs. C. Le Stanc ; RTD com. 2015, p. 307, chron. P. Gaudrat ; Légipresse 2015, p. 223, note N. Binctin ; Comm. comm. électr. 2015, n° 19, obs. C. Caron ; Propr. industr. 2015, n° 3, comm. 25, obs. N. Bouche. Some authors are nevertheless considering the opportunity to vest a legal person with the authorship. On the topic, see A. Bensamoun, La personne morale en droit d’auteur : auteur contre-nature ou titulaire naturel ?, D. 2013, p. 376.
[35] P.B. Hugenholtz/J.P. Quintais (2021), Copyright and Artificial Creation: Does EU Copyright Law Protect AI-Assisted Output?, International Review of Intellectual Property and Competition Law 52 (2021), 1190. The authors underline the presence of the human intellectual effort in the European law and distinguish the criteria from the originality condition, see also, J. C. Ginsburg, People Not Machines: Authorship and What It Means in the Berne Convention, 49 International Review of Intellectual Property and Competition Law, no. 2, 131 (2018).
[36] CJUE, 1st Decembre 2011, Painer, C‐145/10, points 87 à 89 ; CJUE, 12 sept. 2019, aff. C-683/17, Cofemel, point 54 : the aesthetic effect shall not be taken into account to assess the criteria of intellectual creation reflecting the freedom of choice and the personality of the author.; CJUE, 29 juill. 2019, aff. C‐469/17, Funke, point. 19: reflection of the personality by the expression of creative skills in the making of the work by free and creative choices.
[37] Since Feist, Feist Publications v. Rural Telephone Service Company, Inc. 499 U.S. 340 (1991), the Supreme Court has relinquished the criteria of the sweat of the brow and courts have considered that a monkey could not be the author of a selfie: Naruto et al. v. D. J Slater et al., Northern District of California, case n° 15-cv-04324-WHO, 18 janv. 2016 ; 6 ; U.S. Court of Appeals for the 9th Circuit, 28 avr. 2018.
[38] Australian Supreme Court, Telstra Corporation Limited v Phone Directories Company Pty Ltd [2010], FCAFC 149.
[39] The fixation criterion has been interpreted broadly and covers all tangible copies and digital files CA Paris, 4ème ch., 16 juin 2000, Com. com. électr. 2000, comm. 126, note Caron ; JCP E 2001, p. 1382, note Lefranc ; C. Cass. Civ. 1ere 11 septembre 2013, SPEDIDAM, n° 12-17.794.
[40] The courts don’t require that the sounds or images fixed in the phonogram are works to be protected by the neighbouring right of the producer (see for birds singing) CA Paris, 4ème ch., 6 octobre 1979, D. 1981, p. 190, note Plaisant, RTD Com. 1980, p. 346 ; obs. Françon. Not only the requirement of an original work does’nt exist but the human factor is not considered neither as regards the person producing the phonogram.
[41] Art. 1 International Convention on the Protection of Performers and Phonogram Producers, Rome 26 October 1961.
[42] The investment criterion doesn’t provide for a proper threshold for accessing to the protection. Generally, courts require the triple demonstration: initiative, liability and financial risk for the first fixation.
[43] Article L. 113-1 of the French Code on Intellectual Property: La qualité d’auteur appartient, sauf preuve contraire, à celui ou à ceux sous le nom de qui l’œuvre est divulguée. / In the absence of proof to the contrary, authorship belongs to the person or persons under whose name the work is disclosed.
[44] On this analogy, v. M. Senftleben, préc. p. 16. “The parallels between the domaine public payant and the proposed AI levy system are striking. Both concepts concern creations that fall outside the scope of the exploitation rights of individual authors: literary and artistic works that never or no longer enjoy copyright protection in the case of the domaine public payant; general ideas, concepts and styles in the case of AI output that does not reproduce individual expression of a human author.”
[45] Revue des Deux Mondes, 15 janvier 1841, Lettre en faveur de Melle Sedaine.
[46] Pierre-Jules Hetzel, La propriété littéraire et le domaine public payant, républicain qui comme l’auteur des Misérables, Bruxelles, 1858, réimprimé à Paris en 1862.
[47] Victor Hugo, « Actes et Paroles IV, Depuis l’exil 1876-1885 », Société d’Editions littéraires et artistiques, Paris, 1929, p. 411
[48] Article L 111-4 French Code on Intellectual Property according to which protection is granted on the reciprocity condition. The amount of money collected is doomed to be used by a specific entity in charge of public interest. Art. R 111-1 et R 111-2 du CPI.
[49] WIPO, Analysis of the replies to the survey of existing provisions for the application of the system of “Domaine Public Payant” in national legislation”, 1982, WIPO/DPP/CE/1/02 ; CPY.82/WS/1. Many countries such as Algeria, Argentina, Brasil, Bulgaria, Hungary, Italiy, Mexico, Portugal, former Tchecoslovaquia, Tunisia, former USSR and Zaïre have had such systems. Pre-cited Report of S. Dusollier for WIPO published 2010 quote Algeria, Kenya, Rwanda, Senegal, Democratic Republic of Congo, Ivory Coast and Paraguay. Two other countries Argentina and Uruguay had such a regime respectively since 1958 and 1937.
[50] E. R. Harvey, Le domaine public payant dans la législation comparée (et plus particulièrement dans la législation argentine), Copyright Bulletin, XXVIII, 4, 1994, p. 31 and sq.
[51] On the Argentina system see M. Marzetti, https://theconversation.com/un-domaine-public-payant-loxymore-propose-par-victor-hugo-197764
[52] Italy has suppressed in 1996 the system of State Domain that has existed since 1925. See L. Chimenti, Lineamenti del nuovo diritto d’autore- Direttive comunitarie e normativa interna, 2. A, Mailand, 1997, s. 198.
[53] Bangui’s Agreements in the revised Bamako version in 2015, article 68 (previously article 59).
[54] C. Mouchet, Les problèmes du domaine public payant, Le droit d’auteur (BIRPI), 1970, p. 208.
[55] For a very comprehensive review, J. Cayron, A. Albarian, Financer la création culturelle par l’instauration d’un domaine public payant : le renouveau contemporain d’une notion ancienne, Legicom 2006/2, n° 36, p. 117-131. Namely quoted numerous comments of the beginning of the XXth century ; J. Vilbois, Du domaine public payant en matière de droit d’auteur, Théorie, Pratique et Législation comparée, Sirey, 1920, p. 84, n° 59 ; E. Morel, Le domaine public payant, Mercure de France, 1er août 1927, p. 513 ; P. Robiquet, Une nouvelle forme de la propriété littéraire ; le droit d’auteur et le domaine public payant, L’économiste français, 3 octobre 1903, p. 473 ; M. Ajam, La réorganisation des droits d’auteur, Revue Politique et Parlementaire, 10 octobre 1907, p. 5 ; A. Lamandé, Le domaine public payant, Revue Bleue, 3 décembre 1927, n° 23, p. 721.
[56] A. Dietz, Le droit de la communauté des auteurs : un concept moderne de domaine public, Copyright bulletin, 1990, XXIV, 4, p. 13-24. A. Dietz, “A concept of « Domaine public Payant”, in the field of the Neighbouring Right of Performers”, Intellectual Property and Information Law, Essays in honor of Herman Cohen Jehovan, The Hague, London, Boston, Kluwer Law International, 1998, p. 121 et s. See more recently, R. Giblin (2017), “Reimaging Copyright’s Duration”, in R. Giblin/K. Weatherall (eds.), What if We Could Reimagine Copyright, Canberra: ANU Press 2017, 177.
[57] M. Blakeney, The protection of traditional knowledge under intellectual property law, European Intellectual Property Review, 2000, p. 257: “The moneys thereby received can be diverted to the promotion of cultural activities. This scheme is particularly suited for the nurturing of traditional works” ; J. Githaiga, “Intellectual property law and the protection of indigenous folklore and knowledge”, Murdoch University Electronic Journal of Law, Vol. 5, number 2, 1998, § 53, Therefore indigenous folkloric works that are in the public domain would generate revenues for the indigenous owners” ; also see S. Von Lewinski, The Protection of Folklore, in Symposium on Traditional knowledge, Intellectual property, and Indigenous culture, 11 Cardozo J. Int’l & Comp (Cardozo Journal of International and Comparative Law) 767 : “Another way to protect the financial interests of indigenous peoples would be to establish a statutory remuneration right in favour of the community which would itself exercise the power to decide how to income from the exploitation of folklore would be allocated. Such protection could be based upon the model of ‘domaine public payant’” ; S. La Voi, Comment : Cultural heritage tug of war : balancing preservation interests and commercial rights, DePaul Law review, 2003, p. 922.; C. Berryman, Toward more universal protection of intangible cultural property, Journal of Intellectual Property Law, 1994, p. 308 : “Domaine public payant is characterized as a protector of cultural heritage because it can provide the financial means for nations to protect and preserve their cultural creations, particularly folklore”.
[58] S. Dusollier, « L’exploitation des œuvres : une notion centrale en droit d’auteur », in Mélanges A. Lucas, LexisNexis, 2014, p. 263, spéc. p. 267.
[59] This option is preferred by M. Senftleben, quoted who consider that only the persons who don’t use the faculty to opt out could benefit from a levy system on the outputs.
Le "chat" et la souris 
Laisser un commentaire